Hjorth Descriptor Measurement on Multidistance Signal Level Difference for Lung Sound Classification

Abstract

Biological signals have a multiscale nature; hence, many multiscale methods for biological signal analysis have been developed. One of the most popular multiscale methods is the coarse-grained procedure. The coarse-grained procedure has some drawbacks, such as a decreased variance of the signal, since the coarse-grained procedure eliminates the fast temporal scale. As such, other multiscale methods were developed to overcome the limitation of the coarse-grained procedure. In this study, we proposed a new multiscale method that preserves variance of the signal. In our proposed method, we split the signal into a new sequencing signal by using the multi-distance signal level difference (MSLD) method. In MSLD, a set of new signals emerged from the absolute value of two data samples' difference at a defined distance. To evaluate the MSLD performance, we used Hjorth descriptor as the feature extraction method in the output signal. The results were classified using multilayer perceptron (MLP). The proposed method was tested on five classes of lung sound data. The results showed that the proposed method achieved the maximum accuracy of 98.76% for the 81 data. The resulting accuracy was higher than the multiscale Hjorth descriptor using the coarse-grained procedure in our previous research. The MSLD could be combined with feature extraction methods other than Hjorth descriptor for future studies